Our new open book is being actively updated, thanks to the great effort by the team. The following thread provides a nice overview of the book and its latest updates. Please always follow Druv Pai and Sam Buchanan for the very latest about this open project.

three years ago, DiT replaced the legacy unet with a transformer-based denoising backbone. we knew the bulky VAEs would be the next to go -- we just waited until we could do it right. today, we introduce Representation Autoencoders (RAE). >> Retire VAEs. Use RAEs. 👇(1/n)

Unpopular opinion: Finding a simple idea that actually works is way harder than publishing a fancy one that kinda works. You have to fight the urge to overcomplicate, give up many fancier ideas, fail and pivot again and again until you hit the first principle that truly holds.

#PhySIC! 🚀 Ever wanted to turn a single photo into a full 3D world where people interact realistically with surrounding? PhySIC does just that, reconstructing both humans and scenes with physically plausible contact. 🔗 Project Page: yuxuan-xue.com/physic #SIGGRAPHAsia2025 #AI

#InfiniHuman: Infinite 3D Human Generation with Precise Control How do you want to generate a 3D avatar? From text description? With clothing images? Or some desired body shape? All can be done at once with InfiniHuman! 🔗Page: yuxuan-xue.com/infini-human/ #SIGGRAPHAsia2025 #AI

See you in Hawaii #ICCV2025

Human history is marked by the machines we created: from the Antikythera mechanism of ancient Greece, to the imaginations of the Renaissance, to the engines of the steam era. We wonder: can LLMs, like humans, build sophisticated machines to achieve purposeful functionality?

Please check out this tool! It enables easy generation of dynamic G-buffers and data under various lighting conditions. It’s used for data generation in DiffusionRenderer, UniRelight, and LuxDiT. 🚀

Rockets once built by humans, now by AIs! 🚀 We introduce BuildArena 🏟️, where LLM agents design, build, and test rockets, cars, and bridges in a physics simulator. This is #AIforEngineering in action! As AI researchers, we've always wondered: could an AI build a machine?

Heading to Hawaii🏝️for #ICCV2025 ? You definitely won't want to miss our workshop! Interactive Human-centric Foundation Models! ⏰Oct 19 9:00 am - 5:15 pm 📍Room 317A We have Michael Black Ziwei Liu Umar Iqbal, Yuan Zhou, Evonne Ng, Ailing Zeng Lan Xu Angjoo Kanazawa

This is the most inspiring work I've encountered recently. Kudos to you, Zhen Liu! The next logical step is to harness VML (Verbalized Machine Learning) to create a suite of machines with diverse capabilities that can work together toward groundbreaking objectives. Imagine

人最圆滑的时候，是人最没有安全感的时候。 一个人有了安全感，他的棱角就都出来了。 ——马东

充分说明哪怕是这个星球最聪明的头脑，也解决不了论文署名排序的问题

充分说明哪怕是这个星球最聪明的头脑，也解决不了论文署名排序的问题

📢 The MUSI workshop is happening TOMORROW (Oct 20) between 1 pm-5:30 pm! 🙋‍♀️🙋Feel free to add any (spicy) questions on multimodal spatial intelligence you want me to ask during the panel discussion at app.sli.do/event/mg2o72SK…

‼️500h of 3D motion data released‼️ Our team at the Codec Avatars Lab just released a large scale dataset of 3D tracked human motion, including audio and text annotations. Check it out here: meta.com/emerging-tech/…

Does RANSAC matter with VGGT? Yes.

SMPL is 10 years old and has done what we hoped — it changed the way the field estimates and models 3D humans and their motion. I’m delighted that the original team has been recognized today at #ICCV2025 with the Mark Everingham Prize. The prize is given to individuals or

3:15 p.m — 5:15 p.m Exhibit Hall I #851

3:15 p.m — 5:15 p.m Exhibit Hall I #767